💭 putting a maximum word limit is a bit like negative reasoning tokens, but can we go farther? Would writing the answer in code or something take additional reasoning power and endumbify further?